reatom-feedback-loop

To correct Reatom guidance, analyze why the earlier instruction failed to constrain behavior, then distill the follow-up correction into preventive skill edits or evaluation prompts. This ensures the agent's scaffold workflow deviation is fixed durably rather than relying on memory adjustments.

Instruction-shape failure in Reatom debugging occurs when an earlier instruction fails to constrain agent behavior, causing process deviations or incorrect outcomes. Analyzing these failures helps identify where to place preventive rules in always-loaded or referenced guidance locations.

To stop repeating Reatom routing errors, convert the failure into concrete, skill-oriented edits like execution gates, examples, or evaluation prompts. This approach places preventive rules in the correct guidance locations instead of blaming agent memory.

Yes, running a Reatom feedback loop requires reading the skill's feedback-loop reference. This reference provides the necessary structure for performing deviation analysis and proposing specific skill text or gate placements to prevent recurrence.

The best way to improve Reatom skill instructions is to prioritize the immediate user fix first, then propose preventive skill edits or eval additions. This approach analyzes the instruction-shape failure and places rules in the right guidance locations rather than making future-tense promises.